UCSBTH–92–31

July 1992

THE ANGULAR DEPENDENCE OF THE

arXiv:astro-ph/9208001v1 4 Aug 1992

THREE-POINT CORRELATION FUNCTION OF THE

COSMINC MICROWAVE BACKGROUND RADATION

AS PREDICTED BY INFLATIONARY COSMOLOGIES

Toby Falk, Raghavan Rangarajan, and Mark Srednicki

Department of Physics, University of California, Santa Barbara, CA 93106

ABSTRACT

Inflationary models predict a definite, model independent, angular dependence

for the three-point correlation function of ∆T /T at large angles (> ◦
∼ 1 ) which we
calculate. The overall amplitude is model dependent and generically unobservably
small, but may be large in some specific models. We compare our results with
other models of nongaussian fluctuations.

Subject headings: cosmology: theory — large-scale structure of universe

 Recent results from COBE (Smoot et al. 1992) have provided strong evidence for
fluctuations ∆T in the cosmic microwave background radiation (CMBR) on large angular
scales (> ◦
∼ 1 ). The two-point correlation function of ∆T is consistent with a Harrison–
Zel’dovich spectrum, as predicted by inflationary cosmologies (Guth 1981). Here we point
out that inflation results in a definite and model independent prediction for the angular
dependence of the three-point function as well. The overall amplitude of the three-point
function, however, is model dependent. It is possible to construct models (discussed below)
where this amplitude would be large enough to be seen in the COBE data, but generically
it is expected to be too small to be observable.
For large angular scales, ∆T is predicted to be (Peebles 1982; Abbott & Wise 1984;
Bond & Efstathiou 1987)
  Z
3 1/2 ~
∆T (â) = Q d3k ξ(~k) eik·â , (1)
5π

where â is a unit vector on the sky, Q is the predicted RMS quadrupole amplitude, and
the two-point correlation function of the random variable ξ(~k) is given by

ξ(~k1 )ξ(~k2) = k1−3 δ 3 (~k1 + ~k2 ) , (2)

where δ 3 (~k) is the three-dimensional Dirac delta function. To study the two-point corre-
lation function of ∆T in a realistic setting, we must first expand ∆T in Legendre polyno-
mials, remove the monopole, dipole, and (following Smoot et al. 1992) quadrupole terms,
and weight the remaining terms to account for the finite beam width σ of the antennas.
This results in (Bond & Efstathiou 1987)
∞
X
~
eik·â → il (2l + 1) Wl jl (k) Pl (k̂·â) , (3)
l=3

where k = |~k|, k̂ = ~k/k, jl (k) is a spherical Bessel function, and Wl ≃ exp[− 12 (l + 12 )2 σ 2 ];

Smoot et al. (1992) take σ = 3.2◦ . This implies that the two-point correlation function for
∆T takes the form
∆T (â)∆T (b̂) = 56 Q2 C2 (â, b̂) , (4)

where
∞
X
C2 (â, b̂) = (2l + 1) Cl Wl2 Pl (â· b̂) , (5)
l=3
R∞
with Cl = 2 0 dk jl2 (k)/k = 1/l(l + 1). For reference, C2 (â, b̂) is plotted as a function of
the separation angle α = cos−1 (â· b̂) in Figure (1).
2
 In inflationary models, the random variable ξ(~k) has its origin in the fluctuations
of a quantum field ϕ. Working in momentum space, and assuming ϕ is massless and
noninteracting, during the de Sitter epoch the two-point correlation function of ϕ is

ϕ(~k1 , τ )ϕ(~k2, τ ) = 4π 3 H 2 k1−3 δ 3 (~k1 + ~k2 ) (6)

in the long wavelength limit (Bunch & Davies 1978). Here τ is conformal time and H is
the Hubble parameter, related to the (constant) scalar potential V0 via H 2 = 8πV0 /3. (We
take h̄ = c = G = 1.) The three-point correlation function vanishes identically. However,
if the field ϕ has any sort of interaction (with itself or with other fields), nontrivial higher-
point correlation functions will appear, and these will be passed on to the temperature
perturbations (Allen, Grinstein, & Wise 1987). The simplest possible interaction, and one
which will generically be present, is a modification of the scalar potential from a constant
to V (ϕ) = V0 + 61 µϕ3 . To leading order in µ, this interaction implies a three-point function
which can be calculated via standard methods of quantum field theory in curved space
(Birrell & Davies 1982). In the long wavelength limit, we find

ϕ(~k1 , τ )ϕ(~k2, τ )ϕ(~k3 , τ ) = 32 π 3 µH 2 (k1 k2 k3 )−3 F (k1 , k2 , k3 )

× δ 3 (~k1 + ~k2 + ~k3 ) , (7)

where
 
F (k1 , k2 , k3) = (k13 + k23 + k33 ) log(kT |τ |) + γ
− (k12 + k22 + k32 )kT + k1 k2 k3 . (8)

Here ki = |~ki | and kT = k1 + k2 + k3 ; γ is Euler’s constant. The factor of log(kT |τ |)

in equation (8) is minus the number of e-foldings from the time at which fluctuations of
wavenumber kT first passed outside the horizon until the end of inflation. For the relevant
length scales, one typically finds log(kT |τ |) ≃ −60 (e.g., Bardeen, Steinhardt, & Turner
1982), and so the term with this factor dominates over the others. From here on we will
approximate F (k1 , k2 , k3 ) as

F (k1 , k2 , k3 ) ≃ −β (k13 + k23 + k33 ) (9)

and treat β ≃ 60 as independent of ki . A similar logarithmic dependence on ki (with a

different origin) is also present in equation (2), and is also routinely ignored.
The fluctuations in ϕ become fluctuations in the CMBR in one of two ways. If the
energy stored in ϕ is eventually converted to ordinary matter and energy (including the
CMBR), then the fluctuations are adiabatic, and the correlation functions of ξ(~k) are the
3
same (up to a sign) as those of ϕ(~k)/(4π 3H 2 )1/2. In this case we typically find µ/H < ∼
10 −7 (Hodges et al. 1991), but larger values may be possible if we fine-tune the scalar
potential. If, instead, the energy stored in ϕ is eventually converted to dark matter, then
the fluctuations are isocurvature (Linde 1984), and the correlation functions of ξ(~k) are
again the same (up to a sign) as those of ϕ(~k)/(4π 3H 2 )1/2 (Allen et al. 1987). In this case
there are no obvious restrictions on µ/H.
Neglecting, for the moment, the effects of finite beam width and the need to subtract
the l = 0, 1, 2 terms in ∆T , we can combine Eqs. (1), (2), (6), (7), and (9) to compute an
“unsubtracted” three-point correlation function for ∆T . We find
 
1 3 3/2 µ 3 e
∆T (â)∆T (b̂)∆T (ĉ) unsub = ε β Q C3 (â, b̂, ĉ) , (10)
π 5 H
where ε = ±1 is a model dependent sign, and
Z
1
e
C3 (â, b̂, ĉ) = d3k1 d3k2 d3k3 δ 3 (~k1 + ~k2 + ~k3 )
3(2π)2
~ ~ ~
× (k1−3 k2−3 + . . .) ei(k1 ·â+k2 ·b̂+k3 ·ĉ)
Z Z
1 3 −3 i~k1 ·(â−ĉ) ~
= 2
d k1 k1 e d3k2 k2−3 eik2 ·(b̂−ĉ) + . . .
3(2π)
h i
= 13 Ce2 (â, ĉ)C
e2 (b̂, ĉ) + C e2 (b̂, â)Ce2 (ĉ, â) + C
e2 (ĉ, b̂)C
e2 (â, b̂) .
(11)
In the last line, we have defined an “unsubtracted” dimensionless two-point correlation
function
Z
1 ~
Ce2 (â, b̂) = d3k k −3 eik·(â−b̂)
2π
∞
X
= (2l + 1) Cl Pl (â· b̂) . (12)
l=0
Equation (11) is a beautifully simple formula, but clearly is only a formal relation, since the
monopole term in C e2 is infinite. Equation (11) must be corrected to remove the l = 0, 1, 2
terms, and to account for finite beam width. This is straightforward, and ultimately yields
 
1 3 3/2 µ 3
∆T (â)∆T (b̂)∆T (ĉ) = ε β Q C3 (â, b̂, ĉ) , (13)
π 5 H
where
∞
X
C3 (â, b̂, ĉ) = 1 (2j + 1)(2k + 1)(2l + 1) (Cj Ck + Ck Cl + Cl Cj )
3
j,k,l=3
× Wj Wk Wl fjkl (â, b̂, ĉ) , (14)

4
and we have defined
Z
1
fjkl (â, b̂, ĉ) = dΩn Pj (â· n̂)Pk (b̂· n̂)Pl (ĉ· n̂) . (15)
4π

Here dΩn denotes integration over the unit vector n̂. If desired, fjkl (â, b̂, ĉ) can be expanded
in spherical harmonics of each of the three unit vectors by making use of standard relations
[for example, eqs. (9.79) and (16.90) of Merzbacher (1970)], but the resulting formula for
the coefficients is unwieldy and will not be presented here.
C3 (â, b̂, ĉ) is plotted in Figures (2) and (3) for an equilateral triangle on the sky:
â· b̂ = b̂·ĉ = ĉ·â = cos α. Surprisingly, for this case it is given to a good approximation by
[C2 (â, b̂)]2 , mimicking the relation between the unsubtracted functions C e2 and C
e3 . Clearly
to check for the presence of a nonzero three-point function in the COBE data, one should
look at the smallest feasible values of α. If a signal is found, the specific prediction of
equation (14) could be checked by looking at other configurations of â, b̂, and ĉ, such as
all three in a line on the sky.
The nonzero three-point function implies that the fluctuations are nongaussian. The
particular form of the three-point function specified by equation (9) also arises (to leading
order in λ) in models in which fluctuations in the gravitational potential φ are given by a
local, nonlinear transformation of a gaussian random field ξ: φ(~x) = ξ(~x) + λξ(~x)2 + . . . .
Such models have been investigated by Kofman et al. (1991), Moscardini et al. (1991), and
Scherrer (1992). In particular, Kofman et al. (1991) have noted that such effects can arise
due to nonlinearities in the classical evolution equations for certain inflationary models,
even if the initial fluctuations are gaussian. Other nongaussian models (which do not result
from inflation) give different three-point functions. For example, Weinberg & Cole (1992)
have studied models in which the density fluctuations δρ/ρ are given by a local, nonlinear
transformation of a gaussian random field; this yields F ∝ (k1 k2 k3 )2 (k1−1 + k2−1 + k3−1 )
to leading order in λ. Still another type of model has been proposed by Scherrer &
Schaefer (1992); it gives F ∝ (k1 k2 k3 )3/2. These results for F imply that, if we normalize
the deviations from gaussian statistics at small scales, inflationary models will produce
larger deviations on large scales than these alternatives. We would argue that future work
on nongaussian models should focus on those which reproduce the three-point function
reported here, since it is a generic prediction of inflationary cosmologies.
We would like to thank Mark Wise for pointing out to us both the feasibility and
potential interest of this calculation, and for providing us a copy of his unpublished notes
on the related calculations reported by Allen et al. This work was supported in part by
NSF Grant No. PHY-91-16964.

5
 REFERENCES

Abbott, L. F., & Wise, M. B. 1984, Phys. Lett., B135, 179

Allen, T. J., Grinstein, B., & Wise, M. B. 1987, Phys. Lett., B197, 66
Bardeen, J. M., Steinhardt, P. J., & Turner, M. S. 1983, Phys. Rev. D, 28, 679
Birrell, N. D., & Davies, P. C. W. 1982, Quantum Fields in Curved Space (Cambridge:
Cambridge University Press)
Bond, J. R., & Efstathiou, G. 1987, MNRAS, 226, 655
Bunch, T. S., & Davies, P. C. W. 1978, Proc. R. Soc. Lon., A360, 117
Guth, A. 1981, Phys. Rev. D, 23, 347
Hodges, H. M., Blumenthal, G. R., Kofman, L. A., & Primack, J. R. 1990, Nucl. Phys.,
B335, 197
Kofman, L., Blumenthal, G. R., Hodges, H. M., & Primack, J. R. 1991, in Large Scale
Structures and Peculiar Motions in the Universe, ed. D. W. Latham & L. A. Nicolaci
da Costa (San Francisco: Astron. Soc. Pac.)
Linde, A. D. 1984, JETP Lett., 40, 1333
Merzbacher, E. 1970, Quantum Mechanics (New York: Wiley)
Moscardini, L., Matarrese, S., Lucchin, F., & Messina, A. 1991, MNRAS, 248, 424
Peebles, P. J. E. 1982, ApJ, 263, L1
Smoot, G. F., et al. 1992, ApJ, in press
Scherrer, R. J. 1992, ApJ, 390, 330
Scherrer, R. J., & Schaefer, R. K. 1992, Ohio State Univ. preprint OSU-TA-7/92
Weinberg, D. H., & Cole, S. 1991, U. C. Berkeley preprint CfPA-TH-91-025

6
 FIGURE CAPTIONS

Fig. 1.—The dimensionless two-point correlation C2 (â, b̂) as a function of α, where â· b̂ =
cos α, for σ = 3.2◦ . (Description: C2 (α) is the same as the center of the gray band in
Fig. 3 of Smoot et al. but with a different vertical scale; C2 (0) = 3.04.)
Fig. 2.—The dimensionless three-point correlation C3 (â, b̂, ĉ) as a function of α, where
â·b̂ = b̂·ĉ = ĉ·â = cos α, for σ = 3.2◦ . (Description: C3 (α) is defined only for α between
0◦ and 120◦ . Over this range it looks a lot like C2 (α)2 ; C3 (0) = 9.15 and C3 (120◦ ) = 0.)
Fig. 3.—Same as Figure 2, with the vertical scale expanded to show the large angle struc-
ture.